casar

Implementation of CaSAR: Contact-aware Skeletal Action Recognition

About

• label_test_data.py
  • A Flask webserver that provides API (load frames, get action sequence, put label) for manual labeling of test dataset
  • Used to generate a labeled for H2O test dataset. Our label results in: action_test.txt
• modules.py
  • CaSAR torch model and dataset definitions
  • test.py
    • CaSAR testing script
• train.py
  • CaSAR training script
  • Training configuration found in config.yaml
• visualize.ipynb
  • Python notebook for visualizing H2O dataset

Setup

1. Install dependencies

   pip install -r requirements.txt
   

   • Implemented on Python 3.11.8
2. Download H2O dataset: https://taeinkwon.com/projects/h2o/
3. Dataset preparation
   • Replace action_labels/action_test.txt (unlabled) in H2O dataset with action_test.txt (labeled) provided in this repository